Machine Learning Basics

What is Machine Learning?

Machine Learning = Learning patterns from data WITHOUT explicit programming

Instead of writing rules by hand, the computer learns rules from examples.

Simple Comparison

Traditional Programming:
  Rules → Computer → Output
  (You write the rules)

Machine Learning:
  Data + Examples → Computer learns → Rules → Output
  (Computer finds the rules)

---

Three Types of Learning

1️⃣ Supervised Learning

Labeled data provided (X, y)

Training Data:
  Input (X) → Output (y)
  Size=1000 → Price=50L
  Size=2000 → Price=80L
  
Model learns: price = f(size)

Examples:

• Predicting house prices (regression)
• Classifying emails as spam (classification)
• Disease diagnosis from symptoms

Algorithms: Linear Regression, Logistic Regression, Decision Trees

---

2️⃣ Unsupervised Learning

No labels provided (X only)

Training Data:
  Input (X) only
  Customer A: {purchase_history, browsing_patterns}
  Customer B: {purchase_history, browsing_patterns}
  
Model learns: Find similar customers

Examples:

• Grouping customers by behavior
• Finding similar products
• Document clustering

Algorithms: K-Means, Hierarchical Clustering, PCA

---

3️⃣ Semi-Supervised Learning

Mix of labeled and unlabeled data

Training Data:
  100 labeled emails: {content} → {spam/not spam}
  900 unlabeled emails: {content} → unknown
  
Model learns: Find patterns from labeled, apply to unlabeled

Examples:

• Spam detection with partial labels
• Image classification with few labeled images
• Text classification with limited labeled data

---

Exam Question Format

Q: Classify these as supervised/unsupervised/semi-supervised

Task	Answer	Reason
Predict house price from features	Supervised	Have labeled (price) examples
Group customers by behavior	Unsupervised	No labels, discover groups
Label some emails, find patterns	Semi-supervised	Mix of labeled + unlabeled
Classify iris flowers	Supervised	Have labeled flower types
Find document topics	Unsupervised	No predefined labels

---

Training vs Testing Data

The Split

Raw Data (100 samples)
    ↓
    ├─ Training Set (80%) → Used to LEARN
    │
    └─ Test Set (20%) → Used to EVALUATE

Why This Matters

Training Error ≠ Test Error

If training error = 0.05 and test error = 0.40
→ OVERFITTING (model memorized training data)

---

Key Concepts for Your Exam

Term	Meaning
Features (X)	Input variables
Target (y)	Output we predict
Model	Learned rules/patterns
Prediction	What model outputs
Residual	Actual - Predicted
Error	How wrong the prediction is
Accuracy	% of correct predictions

---

Common Exam Questions

Q: Why split data into training and testing?

A:

• Training data: Teach the model
• Testing data: Evaluate how well it generalizes
• Prevents measuring on data the model already saw

Q: What's the relationship between training and test error?

A:

• Good model: Both low and similar
• Overfitting: Train low, test high (gap is large)
• Underfitting: Both high (model too simple)

---

Next Steps

• Move to Linear Regression to learn the main Module 1 topic
• Or review Machine Learning Types here first
• See Practice Problem 6.1 for a complete example

---

Key Takeaways

✅ ML learns patterns from data, not hand-coded rules
✅ Three types: Supervised (labeled), Unsupervised (unlabeled), Semi-supervised (both)
✅ Always split data: Training (learn) + Testing (evaluate)
✅ Good model generalizes: training error ≈ test error
✅ Bad model overfits: train error << test error